# Acceptable Use

CantHarm is intended for research, audit, benchmark reproduction, and analysis of Cantonese harmful lexical forms, linked senses, and controlled canonical audio.

## Acceptable Uses

- dataset inspection and documentation review
- form-level and sense-level harmfulness classification
- severity prediction research
- audio-only, ASR cascade, multimodal, and retrieval experiments within the documented benchmark scope
- scholarly citation and reproducibility studies

## Unacceptable Uses

- speaker recognition, biometric modeling, voice cloning, or re-identification
- production moderation without independent validation and human oversight
- punitive decisions about real people or groups
- demographic inference from speaker audio
- redistribution of dictionary-derived definitions/source text without source-specific permission

## Contact

Questions, takedown requests, or safety concerns: yueyu_dimsum@163.com. Backup: qijiayin@139.com.

Status: public release acceptable-use guidance.